Developer skills
Use the repository-local skills for the relevant development task:
| Task | Skill |
|---|---|
| Assess a design or refactor | design-review traces callers, interfaces, state ownership, and evidence to recommend scoped improvements. |
| Develop a repository change | local-dev requires proportional verification and flow docs for non-trivial runtime changes. |
| Write or review technical docs | technical-writing covers source-backed prose, runnable instructions, page selection, and specification clarity. |
| Write or audit tests | test-audit checks observable behavior, credible regressions, distinct coverage, and production seams. |
| Choose validation or diagnose CI | enterprise-testing routes to the existing testing procedures and exact run/job evidence. |
| Run optional end-to-end acceptance | oceinteg runs named real Helm scenarios only when the user explicitly invokes oceinteg <scenario> or $oceinteg <scenario>; never invoke it automatically for development or general testing requests. |
| Explain a change with a diagram | mermaid-diagrams provides a compact Mermaid template, semantic colors, and honest implementation boundaries. |
| Clean the current diff | deslop permits only behavior-neutral cleanup before independent review. |
| Run requested independent review | autoreview owns the reviewer CLI, isolation, and result interpretation. |
These skills are checked into .agents/skills; no global installation is needed.
Testing setup and real-runtime requirements remain owned by the
testing guides. Each skill describes its scope and prerequisites.
Provenance and updates
design-review is maintained in this repository. The
design philosophy owns its rationale;
Readable code owns the examples. Keep the skill
focused on the review procedure. When changing it, check those references and
try a bounded review task against real source and callers.
technical-writing adapts Docy's references/core/main.md (document lifecycle
and universal technical writing), references/ref/developer-docs.md,
references/ref/concise-instructions.md, and references/ref/spec.md, inspected
on 2026-09-16. It retains writing, source-evidence, example, review, and design
clarity rules while leaving repository policy in AGENTS.md. It omits the Docy
CLI, personal paths, unrelated coding rules, and framework-specific material.
Update its skill and specification reference together; verify local links,
frontmatter, documentation checks, and alignment with repository instructions.
No global Docy installation is required.
local-dev is maintained here. Its flow workflow, template, and standalone
Python validator adapt Specy 2.0.0 (SKILL.md, references/flow-doc/workflow.md,
references/flow-doc/template.md, and scripts/validate_flow_doc.py), inspected
on 2026-09-16. The adaptation replaces personal memory/session/diagram tooling
with repository paths, host-provided provenance, local diagram guidance, and a
bundled standard-library validator. No global Specy installation is required.
Update those resources together; exercise the validator against a completed
flow and malformed input, and verify links and the trivial-change exemption.
The validator checks structure, not source accuracy or Mermaid syntax.
mermaid-diagrams is maintained in this repository. Update its instructions and
template together; check source accuracy and inspect a rendered example when
possible. Report syntax checks and visual inspection separately.
The three adapted skills originate from openclaw/openclaw at commit
4490500902033a1673aed8f42299c232d4b5696f. Their source SKILL.md files are
identical to the audit snapshot at 083b498270124a059db70714b5df93d973391ee0.
The upstream MIT license is retained.
| Upstream source | Intentional Enterprise adaptation |
|---|---|
| test-audit | Retains the four authoring gates and evidence required before deletion. Uses Node conformance/integration tests and Enterprise database/runtime guides; removes OpenClaw wrappers, remote infrastructure, and PR tooling. Independent review follows the requested workflow. |
| deslop | Preserves diff-only, behavior-neutral cleanup before review; links the readable-code guide for relevant examples. Omits the Oxlint claim and explicitly retains fail-closed checks, deferred-work TODOs, and integration-test intent comments. |
| openclaw-testing | Renamed enterprise-testing; replaces commands and specialized routes with Enterprise guides. Retains proportional proof and exact CI diagnosis without importing OpenClaw release, package, or remote infrastructure. |
Update adaptations by comparing the pinned upstream files with a newly selected commit, then applying relevant changes against current Enterprise commands and instructions. Update this table and commit together. Do not overwrite them with an upstream directory sync. Check skill frontmatter, local links, named commands, and the documentation checks; review example tasks against the test integrity and runtime boundaries before publishing.
Autoreview has a separate canonical source and must remain an unmodified complete copy; follow its provenance and sync procedure.